自动定位彩色图像中的文本

Automatic text location in color images

提出一种适用于万维网搜索、彩色图像索引和视频索引等多种应用的彩色图像文本定位方法.算法采用多尺度图像分割方法把彩色图像分割为一系列可能包含文本的图像块,用连通区域方法确定每一个文本图像块中文本的精确位置和大小.采用多颜色图像分解方法提取具有相似颜色的连通区域,通过聚类相邻的连通区域得到文本的位置和范围.利用一个图像数据库来测试本文算法,试验表明算法具有较高的正确率.

This work proposed a novel text location method that is suitable for World Wide Web search, color image indexing and video indexing. Image was first segmented into several blocks using multi-scale segmentation approach; then, texts in blocks were localized using connected component algorithm. A multi-color image decomposition method was utilized to extract connected components with uniform color. The location and range of text were obtained by clustering the adjacency connected components. The algorithm was applied to a test image database and showed promising results.